A consulting perspective · Organizational design and development

Designing Business Organizations in the AI Era

From departmental structures to systems where people and AI create value together

  • Edited 20 October 2026
AuthorPhillip LeConsultant in Organizational Restructuring and Development in the AI Era.
Article overview

AI is already part of everyday work for many people. Yet in many businesses, this capability still has no clearly defined place in the management system. Moving from an informal colleague with no recognized role to an established organizational capability requires deliberate design, so that individual gains become shared performance improvements.

Organizational designPeople and AIEPC contractorsKnowledge management
Article contents
01Section

AI is already at work. Does it have a defined role in the organization?

One employee asks AI what an English word means; another uses it to revise an email, develop design ideas or draft an article. Managers use AI to consolidate reports, analyze information and prepare business options. Adoption varies, but AI is becoming increasingly visible at work. In many cases, the change starts with individual needs before the business establishes a shared way of working.

AI therefore resembles a versatile colleague, ready to help and diligently carry out assigned tasks, yet without a recognized role. Many people use it, while ownership, permitted data, task boundaries and responsibility for checking results remain unclear. This support needs to become an organized capability with objectives, quality standards and accountable owners.

This is where the gap emerges between having people who use AI and having an organizational AI capability.

In everyday terms, everyone is doing their own thing. Individuals select tools, save prompts and accumulate experience separately. The 2024 Work Trend Index from Microsoft and LinkedIn reported that 78% of surveyed AI users brought their own AI tools to work. This supports the trend toward individual adoption; it does not mean that every employee or organization has adopted AI. [7]

When use remains fragmented, benefits can stop at individual desks: one person saves time while another still re-enters the data; a useful answer never becomes shared knowledge; a tool gains widespread use without anyone accountable for its impact or risks. This is where the gap emerges between having people who use AI and having an organizational AI capability.

In this article, giving AI a defined organizational role means establishing its purpose, objectives, permitted scope, access rights and accountability arrangements. This approach aligns with NIST principles on clarifying roles, responsibilities and communication lines in AI risk governance. [8] Before creating a new department, a business needs to ask: Why is this capability needed? Whom does it serve? What outcomes should it produce? Who operates it, and who sets its limits?

TRADITIONAL STRUCTUREREDESIGNAI-ENABLED ORGANIZATIONSpecialist departmentsMultiple handoffsKnowledge held by individualsMore work often meansmore peoplePeople and AI withinshared workflowsClear authority and accountabilityValidated knowledge,shared across the businessGreater capacity throughwork redesignWorkDataDecision rightsControlsFrom customer needs to measurable business outcomes
Figure 1. Work and governance arrangements need to change together

The proposed mission is to turn technology, data and business knowledge into a capability that helps people work effectively, serve customers better and expand the business. The lead unit could be a Technology, Data and AI Department or a coordinating team suited to the scale of the organization. AI is not granted legal personality or human accountability; responsibility rests with the people who manage and use the systems.

Organizational design must therefore establish which work AI should handle, which knowledge needs standardization, which decisions require people and which results should be measured. A meaningful role must come with a clear mission, authority and responsibilities, enabling AI to become a capability that the organization develops and controls over time.

02Section

EPC contractors: from a traditional matrix to an AI-enabled matrix

The two illustrative diagrams below retain both management dimensions: functional units provide expertise and resources, while project teams coordinate delivery against schedules, costs and delegated authority. This is a hypothetical model for an Engineering, Procurement and Construction (EPC) contractor, not a verified structure of a particular company.

TRADITIONAL EPC MATRIXEXECUTIVE MANAGEMENTENGINEERINGDesign expertisePROCUREMENTSourcing expertiseCONSTRUCTIONDelivery expertisePROJECTA TEAMProject A · EProject A · PProject A · CPROJECTB TEAMProject B · EProject B · PProject B · CTRADITIONAL ITInfrastructure, ERP, accounts and supportQHSE · Finance · HR support the projectsSolid: management / coordination · Gray dashed: expertiseAI-ENABLED EPC MATRIXEXECUTIVE MANAGEMENTENGINEERINGDesign expertisePROCUREMENTSourcing expertiseCONSTRUCTIONDelivery expertisePROJECTA TEAMProject A · EPeople + AIProject A · PPeople + AIProject A · CPeople + AIPROJECTB TEAMProject B · EPeople + AIProject B · PPeople + AIProject B · CPeople + AITECHNOLOGY, DATA AND AI DEPARTMENTIT + Knowledge + Skills + AI coordinationQHSE · Finance · HR support the projectsTeal dashed: shared platforms, data and AI supportSimplified for comparison; support units and detailed relationships are condensed.
Figure 2. Two EPC models with the same functional and project layout

In the traditional model, IT maintains systems, user accounts, ERP and technical support. In the AI-enabled model, this capability remains within a Technology, Data and AI Department or works alongside a separate AI Department. The expanded capability includes standardizing knowledge, packaging reusable skills, integrating data, testing, and embedding AI assistants or agents in the workflows of each function and project.

Shorter turnaround comes from reducing search, re-entry and information transfers while retaining professional control points.

The AI lead unit reports to executive management so it can coordinate across departments. Its support does not create an additional line of command on site. Functional heads remain accountable for professional standards, and project directors remain accountable for delivery within delegated authority. An authorized person must resolve resource-priority conflicts, budget limits and exceptions.

Consider a technical clarification request: the project team receives the request; AI retrieves the correct document versions and drafts a comparison; an engineer checks it; an authorized person approves it; and the system records and distributes the response. Shorter turnaround comes from reducing search, re-entry and information transfers while retaining professional control points.

In procurement, AI can help compare quotations against consistent technical and commercial conditions. In construction, it can consolidate site logs, track documentation and flag discrepancies. Proposals must include the project code, work package, data sources and update time. Design approval, subcontractor selection, safety decisions and acceptance remain with authorized people. CIOB emphasizes context, experience and human oversight when using AI in construction. [10]

The paired diagrams show where the new capability sits. Implementation also requires an authority matrix, functional charters, data standards and process maps. A simpler chart creates value only when people understand what they may do, whom they must coordinate with and how to handle exceptions.

An illustrative scenario: resources, costs and competitiveness

Assume an EPC contractor maintains the same workload and complexity, with 750 office and project-management staff. Site workers and subcontractor labor are excluded. The post-AI scenario assumes stable operation with validated quality and controls. All figures and scores in this section are hypothetical discussion inputs, not forecasts or measured results.

Hypothetical scenarioIllustrative discussion figures, not a forecast
−9.5%Staff within scope750 → 679 people
−4.7%Monthly cost within the modelVND 18.75 → 17.87 billion
≈VND 10.6bnAnnual net savingsVND 884 million per month
−50%Routine document cycle48 → 24 hours
+9 ppLabor hours for customers15% → 24%
−40%Resource redeployment10 → 6 days
FactorTraditionalAI-enabledAssumed change
Staff within scope750 people679 people71 fewer people / 9.5%
Monthly cost within the modelVND 18.75 billionVND 17.87 billionDown VND 884 million / 4.7%
Annual net savings––Approximately VND 10.6 billion
Routine document cycle48 hours24 hours50% less time
Overall competitiveness5/107/10Qualitative assessment
Labor hours for customers15%24%Up 9 percentage points
Resource redeployment10 days6 days40% less time
CAPABILITY COMPARISON ON A 10-POINT SCALEHypothetical scenario · Higher scores indicate better outcomes on the selected criteria0246810Lean staffing57.8+2.8Cost efficiency57.4+2.4Processing and approval speed59+4Competitiveness57+2Customer-facingcapacity58+3Redeployment flexibility59+4TraditionalAI-enabledChange from baseline
Figure 3. Illustrative scores against target benchmarks; not a ranking of businesses
Assumptions and calculationsRead alongside the table and Figure 3

Assumed staffing structure

Office and project-management staff are grouped by department. Reduction rates reflect assumed AI support for each department’s work (for example, an HR team of 10 becomes 8, an administrative team of 7 becomes 6, and BIM and engineering staffing falls by more than 20%). Reductions across departments total 97 people. After adding 8 business-development staff and 18 IT, data and AI staff, the net reduction is 71, from 750 to 679. This could be achieved through lower replacement hiring or work reassignment, subject to validated workloads and quality.

DepartmentBeforeAssumed rateAfter
Design, BIM and engineering170−22%133
Project and construction management (office)190−10%171
Procurement and contracts70−18%57
Finance and accounting60−20%48
Human resources30−20%24
Administration, document control and coordination50−14%43
QHSE and legal40−8%37
Business development, bidding and customer relations80+10%88
IT, data and AI30+60%48
Executive management and support300%30
Total750−9.5%679

Cost calculation

Assume an average employment cost of VND 25 million per person per month. Before AI: 750 × 25 = VND 18,750 million. After AI: 679 × 25 + 890.6 = VND 17,865.6 million. Approximately VND 890.6 million is incremental monthly AI cost: VND 1.25 million for each person in the pre-AI workforce (VND 937.5 million), less a 5% volume discount. This excludes staff already counted above. The model does not represent total EPC costs and excludes initial investment, transition costs, taxes and salary changes. Staffing benefits must not be counted again as benefits from redeploying the same labor hours.

Score conversion

Baseline = 5; discussion target = 10. Score = 5 + 5 × (result − baseline) / (target − baseline), bounded to 0–10. The respective targets are: 625 people; VND 16.875 billion per month; 18 hours; 30% of labor hours devoted to customers; and 5 days for redeployment. Lean staffing scores 7.8 and cost efficiency 7.4. Overall competitiveness is a separate qualitative assumption of 5→7, requiring validation through quality, schedule reliability, responsiveness and customer feedback. No composite score is calculated.

Three scenarios

The conservative scenario uses lower reduction rates (for example, 15% for design and BIM). The optimistic scenario uses higher rates (25%) and a 10% AI price discount. AI costs determine the outcome: in the conservative scenario, AI costs almost entirely offset staffing savings.

ConservativeBase caseOptimistic
Post-AI staffing710679654
Net reduction5.3%9.5%12.8%
AI volume discount5%5%10%
Monthly net savingsVND 0.11bnVND 0.88bnVND 1.56bn
Annual net savingsVND 1.3bnVND 10.6bnVND 18.7bn

Measuring time

The document cycle runs from receipt of complete information to routine approval, including waiting time. Compare documents of similar complexity. Resource redeployment runs from a priority decision until staff, access rights and skills are ready for the project. A reusable skill library and standardized roles support the move from bidding to delivery and changes in project scale. Benchmarks must be reset using actual data while maintaining quality and safety.

Indirect benefitsNot included in the savings table

Fast, accurate information shortens decision cycles; a lean organization reduces duplication and management effort. These benefits are excluded from the table above to avoid counting released labor hours twice. Businesses can track them through the following indicators.

Fast, accurate information

  • Time to find the correct document version
  • Quotation, RFI and document-approval cycles
  • Time from request to decision
  • Error and rework rates

A leaner organization

  • Reporting layers and spans of control
  • Handoffs and coordination points per process
  • Share of labor hours spent on coordination and administration

Management capacity

  • Weekly meeting hours and overtime
  • Coordination requests per person
  • Management time spent on exceptions

Project and workforce outcomes

  • Schedule delay days and additional costs
  • Bid win rate and revenue per employee
  • Employee turnover
03Section

AI, IT and operations in the new structure

AI and IT manage different domains, but they depend on each other. IT provides a stable, secure technical environment. The AI unit works with process owners to turn knowledge and experience into practical support capabilities. Operating units use both capabilities to deliver results. The three-layer model below describes collaboration; it does not place the AI Department above IT or other departments.

EXECUTIVE MANAGEMENT SETS DIRECTION AND REMAINS ACCOUNTABLEAIKNOWLEDGE AND SKILLSCoordination within delegated scopeIT FOUNDATIONInfrastructure · ERP / BIM · DataIdentity · APIs · Security · BackupsOPERATIONSEngineering · Procurement · Construction · Customer servicePeople and AI execute, check and handle exceptionsTOP-DOWNObjectives and rulesAccess and deploymentBOTTOM-UPData and resultsFeedback for improvementThree collaborating capability layers, not a hierarchy between AI, IT and other departments.Feedback is validated and approved before knowledge or skills are updated.
Figure 4. AI, IT and operations: three capability layers with deployment and feedback flows

Giving AI data to retrieve does not necessarily require retraining the model.

The AI and knowledge layer standardizes business knowledge, rules, output templates and operating limits; packages reusable skills; coordinates assistants or agents within delegated scope; and evaluates quality. The AI Department manages knowledge applications, while experts and process owners validate content accuracy. “Directing information flows” means supporting coordination under approved permissions, not independently setting management rules.

The IT layer manages devices, networks, software, ERP, transactional data, user accounts, integrations, backups and security. AI uses controlled connections to work with ERP or Building Information Modeling (BIM) systems. Transaction data retains an authoritative source, and documents retain version and access controls. Giving AI data to retrieve does not necessarily require retraining the model.

The operations layer includes engineering, procurement, construction and business-support teams. Staff assign work, check outputs and handle exceptions. Department heads and project directors remain accountable for results within their delegated authority. Processes must specify what AI may execute, what it may only recommend and where human approval is required.

From the top down, executive objectives and policies become processes, skills, configurations and access rights for operational use. From the bottom up, results, errors, exceptions and customer feedback are captured, checked and used to propose improvements. Knowledge and skills are updated only after business validation, testing and version approval; AI is not assumed to learn automatically from all incoming data.

A “skill” is a reusable capability package: purpose, conditions of use, permitted data, instructions, tools, output templates, quality criteria, an owner and a test suite. Executives and experts contribute experience; the AI team standardizes and builds the capability; IT provides integration, permissions and operation; and the process owner accepts it before use.

A small business may combine these capabilities in a Technology, Data and AI Department. A large business may maintain separate IT and AI departments under a shared executive prioritization mechanism. IT is responsible for services and technical controls; AI for AI products and evaluation; process owners for business value. Joint planning, budgets and measures help prevent duplication and responsibility being passed between departments.

04Section

People and AI working together in everyday operations

In this article, “cowork” means people and AI jointly carrying out a process, rather than a particular product. AI assistants support users directly; AI agents can use tools to perform a sequence of steps within delegated limits. Both require clear objectives, appropriate data and a way to validate results.

CUSTOMERNeeds and context1EMPLOYEEConfirm and assign2AI + ERP / CRMRetrieve and draft3APPROVERReview and approve4EXECUTIONSend, log, measure5SHARED FOUNDATIONAuthorized data · Approved skills · Logs · Monitoring · Stop controlsExecutive management sets objectives and limits, and monitors end-to-end results
Figure 5. A proposed collaboration flow from customer request to quotation

Appearing on an organizational chart does not make AI a management layer with independent legal accountability.

For example, when a purchase enquiry arrives, a sales employee confirms the need and asks AI to prepare an option. AI reads authorized documents, retrieves prices and inventory from systems, drafts a quotation and identifies missing information. The employee checks the commercial terms; a manager approves anything above delegated limits; only then does the system send the document and record the transaction. If data conflicts or a request exceeds the permitted scope, AI must escalate to the responsible person rather than complete it independently.

This collaboration changes job descriptions. Employees spend less time consolidating information and more time understanding customers and resolving exceptions. Managers focus on priorities, coaching and quality control. The AI team maintains shared capabilities. Every agent needs a clear owner, task scope and permissions. Appearing on an organizational chart does not make AI a management layer with independent legal accountability.

05Section

Challenges that must be addressed in the design

The first challenge is mindset and motivation. Employees may hesitate to share experience if they believe every improvement will immediately lead to job cuts. Leaders need to explain objectives transparently, provide retraining, assign suitable work and recognize contributions to standardizing knowledge. They must also preserve learning opportunities for new staff. Complete dependence on AI drafts can weaken professional expertise and critical judgment.

The next challenge is accountability and concentrated authority. If every request waits for the AI team, the new unit becomes a bottleneck. Centralizing standards, platforms and controls must be distinguished from empowering departments to apply AI. Process owners remain accountable for business operations; the AI team is responsible for technical quality and operation; control functions retain independent assessment rights proportionate to risk.

Content from external emails, documents or websites must be treated as data to assess, not as instructions that can independently change system permissions.

Data security must be designed into each workflow. Significant risks include disclosure of customer information or trade secrets; unauthorized data retrieval; incorrect outputs trusted as accurate; malicious instructions embedded in external documents to manipulate AI; agents granted excessive powers; and unreviewed changes to data or skills. OWASP lists information disclosure, prompt injection and excessive agency among the major risks for language-model applications. [5]

Proposed measures include classifying data, using approved enterprise accounts, applying least-privilege access to people and agents, checking permissions on every retrieval, managing access keys and keeping logs. Content from external emails, documents or websites must be treated as data to assess, not as instructions that can independently change system permissions. Payments, changes to recipient bank details, contractual commitments and sensitive-data exports require appropriate approval points.

Businesses also need periodic testing, drift monitoring, spending limits, stop controls, recovery plans and incident-response procedures. Vendor policies must clarify data retention, data use, subprocessors, and options for exporting or deleting data. NIST’s Generative AI Risk Management Profile provides a reference for organizing risk identification, measurement and management. Each implementation still requires review of the legal obligations applicable to the business. [6]

06Section

From a defined role for AI to organizational competitiveness

A tool’s ability to act does not automatically give it permission to act.

McKinsey proposes adding a map of human–AI agent collaboration alongside the reporting chart, while clarifying decision rights. [1] BCG emphasizes redesigning end-to-end processes. [2] The consulting implication is that the organizational chart identifies accountability, while the work map explains how people, AI and data collaborate to deliver results.

Start with four questions for each task: Is it still necessary? Can software rules handle it? Is AI suitable? What role must people retain? When using AI, distinguish three levels: analytical support; recommendations for human approval; and execution within delegated limits. New tasks, incomplete data or consequences that are difficult to reverse require greater oversight. A tool’s ability to act does not automatically give it permission to act.

Economics must be assessed across the entire process. Net benefits comprise realizable savings and incremental gross profit, less technology, data, integration, control, training and error-resolution costs. Released labor hours must translate into measurable reductions in overtime or new hiring, or increased service capacity. Do not count both staffing savings and revenue generated from the same released time. NBER research on the productivity J-curve emphasizes complementary investment in processes, business models and human capital. [9]

The 14% productivity gain in one customer-support study [3] is not an EPC staffing-reduction benchmark. EPC businesses need to measure document turnaround and rework costs while checking quality and safety.

Restructuring must preserve the ability to learn. When AI takes on basic tasks, junior engineers and specialists still need guided practice to develop judgment. Managers should focus on priorities, coaching and exceptions, with spans of control based on actual workload. Reducing staff while overloading reviewers can increase risk.

An implementation roadmap with conditions for scaling

  1. 1

    Establish the foundation

    Business objectives, process owners, data quality and baseline measures.

  2. 2

    Run a focused pilot

    One process, covering routine cases and exceptions, over approximately 90 days.

  3. 3

    Scale when ready

    Required quality, verified access, accountable operators and validated economics.

  4. 4

    Adjust the organization

    Tasks, staffing and management layers; decide whether to prioritize growth or cost reduction.

First, establish business objectives, process owners, data quality and baseline measures. Then pilot a narrowly scoped process, test routine cases and exceptions, train users and capture the full costs. Scale only when quality requirements are met, access rights are verified, someone is accountable for operation, and the economics are validated. A pilot may run for approximately 90 days, but scaling decisions should follow evidence, not the calendar alone.

Only after the pilot should tasks, staffing and management layers be adjusted, alongside a decision on whether released resources will support growth or cost reduction. ILO and NASK estimate that roughly one quarter of global employment has some exposure to generative AI, while assessing task transformation as more likely than complete replacement. This indicates potential exposure, not a forecast of job losses. [4]

Evera’s approach to AI-enabled organizational consulting

Evera aims to support businesses from value-stream diagnosis through the design of work, structures, decision rights and Technology, Data and AI capabilities. Consulting should connect the people-and-AI model with functional charters, standard processes, a skill library, data permissions, workforce plans and measurement arrangements. Value is confirmed when the model operates and produces results beyond a tool demonstration.

AI’s new organizational role is therefore that of a governed business capability. AI is expanding the potential to research, design, learn and serve customers. Advantage belongs to organizations that turn that potential into reliable ways of working and continuous improvement. This underpins Evera’s consulting approach: helping businesses build lean organizations, sustain accountability and pursue business objectives that were previously difficult to achieve.

RESEARCH SOURCES

References

The article’s research sources were accessed on 19 October 2026. Models, diagrams and the roadmap are analytical proposals; figures in the EPC example are explicitly hypothetical. Select a citation number in the article to jump to the corresponding source.

  1. 01
    McKinsey. AI Is Changing Work. Now It Has to Change the Organization. 5 October 2026.https://www.mckinsey.com/capabilities/people-and-organization/our-insights/ai-is-changing-work-now-it-has-to-change-the-organization
  2. 02
    BCG. Scaling AI Requires New Processes, Not Just New Tools. 20 January 2026.https://www.bcg.com/publications/2026/scaling-ai-requires-new-processes-not-just-new-tools
  3. 03
  4. 04
    ILO and NASK. Generative AI and Jobs: A 2025 Update. 20 May 2025.https://www.ilo.org/publications/generative-ai-and-jobs-2025-update
  5. 05
    OWASP. Top 10 for LLM Applications 2025. 2025.https://genai.owasp.org/llm-top-10/
  6. 06
    NIST. Generative AI Profile, NIST AI 600-1. 2024.https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-generative-artificial-intelligence
  7. 07
    Microsoft and LinkedIn. Work Trend Index 2024. 8 May 2024.https://www.microsoft.com/en-us/worklab/work-trend-index/ai-at-work-is-here-now-comes-the-hard-part
  8. 08
    NIST. AI RMF Core, Govern 2 and Govern 2.1. 2023.https://airc.nist.gov/airmf-resources/airmf/5-sec-core/
  9. 09
  10. 10
    CIOB. Practical AI for Construction. 2025.https://www.ciob.org/blog/practical-ai-for-construction

Next steps

From an informal AI colleague to an established organizational capability

Every business starts from a different position. The first step is often small but decisive: understand where AI is already used, who is accountable and where the boundaries lie.

Four questions to ask your organization

  • Which tasks already use AI, and who uses it?
  • Who is accountable for data, quality and outcomes?
  • Where may AI only recommend, and where is human approval required?
  • Which results will show whether AI is creating real value?

Any unanswered question is a design priority.

  1. 1

    Diagnose

    Current value streams, work, data and business risks.

  2. 2

    Design

    AI placement, organizational charts, functional charters, role lists and decision rights.

  3. 3

    Pilot with controls

    One focused process over approximately 90 days; scale only when the evidence supports it.

Phillip LeEvera · Organizational Restructuring and Development Consulting in the AI Era
Emailphillip@everaconsulting.vn
Emailphilliphcm@gmail.com
Phone+84 988 578 201